Nextdoor engineers use Codex to build features end-to-end

Nextdoor's adoption of Codex illustrates a shift in how software teams measure their work. Dolphin describes it as moving from iteratively prompting an agent to outcome engineering, where engineers think about the result they want and guide an agent to achieve it. This is significant because it suggests the core job of an engineer is changing from writing every line of code to defining problems clearly enough for an agent to solve.

The company uses Codex not just for new features but for investigation work, like figuring out why Kubernetes pods won't start or finding the right trend line in data analysis. Dolphin credits GPT-5.4 and 5.5 upgrades with making Codex "extremely persistent" in tracing esoteric technical details to their root cause.

For Nextdoor, which serves over 110 million users across 11 countries, this means product decisions that once died in the backlog can now see the light of day. The map feature for Opportunity Alerts is a case in point. The risk is whether this speed leads to better products or merely faster output. The answer likely hinges on whether engineers can maintain quality while delegating more hands-on work to agents.